Diagnose a sudden drop in customer satisfaction for low-bandwidth collaboration mode

Problem Statement Description

You are investigating a sudden drop in customer satisfaction tied to support agents using a low-bandwidth collaboration mode. This mode is intended to help agents continue working with customers and teammates when network conditions are poor, such as during screen sharing, internal escalation, co-browsing, chat handoff, or lightweight voice/video collaboration.

The issue is recent and appears concentrated among support workflows that rely on this mode. Before recommending fixes, your task is to structure a root-cause analysis that separates a real customer-impacting problem from measurement noise, identifies where in the agent/customer workflow the degradation is happening, and determines whether the cause is product, infrastructure, operational, or behavioral.

You should assume this is a live production environment with global users, varied network quality, multiple support channels, and high sensitivity to customer trust. The diagnosis should be practical enough to guide engineering, support operations, and product teams toward the right next steps.

The experience should consider:

- How to define the anomaly: CSAT metric, baseline, time window, affected geographies, support queues, customer types, and agent cohorts.

- Whether the drop is isolated to low-bandwidth collaboration mode or also visible in normal-bandwidth sessions, non-collaboration workflows, or specific support channels.

- Instrumentation checks, including CSAT collection changes, missing events, survey response bias, session tagging accuracy, and recent analytics pipeline changes.

- Funnel and workflow segmentation, such as session start, connection degradation, fallback activation, escalation, handoff, resolution, and post-interaction survey completion.

- Product and technical hypotheses, including latency, dropped connections, degraded audio/video/chat quality, sync issues, permission failures, UI confusion, or unexpected mode activation.

- Operational hypotheses, such as new agent training, staffing changes, queue mix shifts, policy updates, scripted responses, or higher-severity customer issues.

- Evidence needed to prioritize causes, including logs, session quality metrics, agent feedback, customer verbatims, release timelines, incident reports, and comparative cohorts.

- Immediate containment, communication, and prevention considerations without jumping prematurely to a permanent fix.

The goal is to demonstrate how you would lead a disciplined RCA: validate the signal, narrow the blast radius, generate and test hypotheses, identify the most likely root cause, and define what evidence would be required before moving into remediation.